AI Incident Copilot

An intelligent agent designed to assist Site Reliability Engineers (SREs) during incidents. This system leverages Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) to analyze logs and metrics, diagnose root causes, and propose remediation plans grounded in your runbooks.

Features

• Automated Context Collection: Fetches logs and metrics for the affected service.
• Intelligent Diagnosis: Analyzes incident data to identify likely root causes with confidence scores.
• RAG-Grounded Remediation: Generates remediation plans strictly based on existing runbooks to ensure safety and compliance.
• Human-in-the-Loop: Includes a mandatory approval step where SREs can review, approve, or reject (with feedback) the proposed plan.
• Iterative Refinement: If a plan is rejected, the agent uses the feedback to re-diagnose and generate a new plan.
• Automated Execution: Once approved, the agent executes the remediation steps automatically.

Architecture

The system is built using LangGraph to manage the stateful workflow:

1. Retrieve Context: Gathers logs, metrics, and relevant runbook sections.
2. Diagnose: Uses an LLM to determine the root cause.
3. Plan: Creates a step-by-step remediation plan.
4. Human Approval: Pauses for user input.
   • Approve: Proceed to execution.
   • Reject: Provide feedback and loop back to diagnosis.
5. Execute: Runs the approved actions.

Getting Started

Prerequisites

• Python 3.10+
• pip
• Access to necessary LLM APIs (configured via .env)

Installation

1. Clone the repository:

   git clone <repository-url>
   cd sre_agent

2. Create and activate a virtual environment:

   python -m venv .venv
   source .venv/bin/activate  # On Windows: .venv\Scripts\activate

3. Install dependencies:

   pip install -r requirements.txt

4. Set up environment variables: Create a .env file in the root directory and add your API keys:

   # Example
   OPENAI_API_KEY=your_api_key_here
   # Add other necessary keys as per 'model.py' and 'tools.py'

Usage

Run the agent from the command line:

python main.py

The agent will start, analyze the mock incident (configured in main.py), and present you with a diagnosis and remediation plan for approval.

Project Structure

• main.py: Entry point for the application.
• graph.py: Defines the LangGraph workflow and nodes.
• agents.py: Contains the logic for Diagnosis and Planning agents (LLM chains).
• tools.py: Tools for fetching logs, metrics, and searching docs.
• state.py: Defines the execution state schema.
• mock_data.py: Mock data for testing the agent without live systems.
• rag.py: RAG implementation details.